Publish final card checksums and reproducibility manifests

LICENSE

@@ -0,0 +1,209 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Original Qwen model:
+   Copyright 2026 Alibaba Cloud
+
+   Fine-tuning and repackaging modifications:
+   Copyright 2026 TokenRhythm
+
+   NeoHorse-1-4B is based on Qwen/Qwen3.5-4B and was modified
+   by TokenRhythm through fine-tuning and repackaging.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -27,10 +27,9 @@ I build and test these releases myself. Your coffee helps keep me going.<br>
 Thank you for supporting this work.
 </div>
 
-> **Status:** Source lock, direct BF16 conversion, and a CPU chat smoke test
-> passed. GPU calibration, the quantization ladder, hold-out evaluation, and
-> final artifact smoke tests are still in progress. Files will be listed only
-> after direct-from-BF16 conversion and runtime validation pass.
+> **Status:** Complete release. The nine GGUF files below were quantized
+> directly from the locked upstream BF16 snapshot, loaded/generated on GPU 0,
+> and checked against held-out next-token fidelity measurements.
 
 ## About NeoHorse-1-4B
 
@@ -40,40 +39,100 @@ See the [official upstream model card](https://huggingface.co/TokenRhythm/NeoHor
 
 These releases change the storage format through GGUF quantization only. No training or fine-tuning was performed here.
 
+## Quantized files
+
+`Q5_K_M` is the balanced starting point from this evaluation. Use `Q6_K` or
+`Q8_0` when fidelity is more important than size; `Q4_K_M` is the smaller
+practical option. `Q3_K_M` and below are included for experimentation, but the
+hold-out results show a substantial fidelity drop. `IQ1_M` and `Q1_0` are
+compatibility-only options here and are not recommended for normal use.
+
+| File | Size (bytes) | Evaluation note |
+|---|---:|---|
+| `NeoHorse-1-4B-Q8_0.gguf` | 4482403104 | closest conventional quant |
+| `NeoHorse-1-4B-Q6_K.gguf` | 3464055872 | high-fidelity option |
+| `NeoHorse-1-4B-Q5_K_M.gguf` | 3074987072 | balanced starting point |
+| `NeoHorse-1-4B-Q4_K_M.gguf` | 2708804672 | smaller practical option |
+| `NeoHorse-1-4B-Q3_K_M.gguf` | 2262064192 | experimental; larger drift |
+| `NeoHorse-1-4B-Q2_K.gguf` | 1915470912 | experimental; large drift |
+| `NeoHorse-1-4B-IQ2_XS.gguf` | 1630594112 | experimental; large drift |
+| `NeoHorse-1-4B-IQ1_M.gguf` | 1426418752 | compatibility-only |
+| `NeoHorse-1-4B-Q1_0.gguf` | 1038189632 | compatibility-only |
+
 ## Fidelity measurements
 
-This section will be completed after a fixed hold-out evaluation against the
-BF16 GGUF reference. Until then, this page makes no quality or “best
-quantization” claim.
+The table averages eight chunks from each of `wiki.test.raw` and
+`wiki.valid.raw`, using a 4,096-token context and the same llama.cpp runtime
+against the BF16 GGUF reference. The average BF16 PPL in this evaluation was
+8.635307. Lower Mean KLD, ΔPPL, and RMS Δp, and higher Top-1 agreement, indicate
+closer next-token behavior to BF16. These are fidelity measurements, not a
+direct percentage of capabilities retained.
+
+| File | Mean KLD | Top-1 vs BF16 | ΔPPL | RMS Δp |
+|---|---:|---:|---:|---:|
+| `NeoHorse-1-4B-Q8_0.gguf` | 0.006291 | 98.065% | +0.192% | 2.041% |
+| `NeoHorse-1-4B-Q6_K.gguf` | 0.010910 | 96.751% | +0.831% | 2.881% |
+| `NeoHorse-1-4B-Q5_K_M.gguf` | 0.032008 | 94.498% | +1.706% | 4.631% |
+| `NeoHorse-1-4B-Q4_K_M.gguf` | 0.055031 | 91.909% | +1.339% | 5.920% |
+| `NeoHorse-1-4B-Q3_K_M.gguf` | 0.160972 | 85.173% | +11.466% | 10.610% |
+| `NeoHorse-1-4B-Q2_K.gguf` | 0.348854 | 76.007% | +12.720% | 16.212% |
+| `NeoHorse-1-4B-IQ2_XS.gguf` | 0.517311 | 70.662% | +32.407% | 20.489% |
+| `NeoHorse-1-4B-IQ1_M.gguf` | 1.303395 | 54.895% | +168.381% | 32.950% |
+| `NeoHorse-1-4B-Q1_0.gguf` | 12.339880 | 0.406% | +18400566.679% | 60.432% |
+
+The fixed hold-out corpus is disjoint from the calibration inputs. See the
+[compact quality summary](reproducibility/quality-summary.tsv) and the
+[reproducibility manifest](reproducibility/manifest.md) for the exact split,
+hashes, and runtime profile.
+
+## Behavior and runtime validation
+
+BF16 baseline probes covered English, Vietnamese, Chinese, coding, plain JSON,
+and the upstream-style tool-call text format. Those probes generated usable
+responses for the tested prompts. The strict llama.cpp `--json-schema` probe
+failed during runtime sampler initialization and is recorded as a runtime
+compatibility limitation, not counted as a model capability result. The
+tool-call probe tested serialization format only; it was not an end-to-end tool
+executor test.
+
+BF16 and every published quantization loaded and generated non-empty output on
+one NVIDIA A10M using the same GPU-0 profile. Low-bit files can pass this
+compatibility gate while still producing poor responses; the fidelity table is
+why Q3 and lower are marked experimental.
 
 ## Quick start
 
-After a published GGUF file appears in this repository:
+For the balanced starting point:
 
 ```bash
 ./llama-cli \
-  -m <published-file>.gguf \
-  --chat-template-file chat_template.jinja \
+  -m NeoHorse-1-4B-Q5_K_M.gguf \
   --jinja \
   --reasoning off \
   -p 'Answer briefly in English: What is GGUF and why is it useful for running language models locally?' \
   -n 128 -c 4096 -ngl 99
 ```
 
-The exact filename, runtime validation profile, and supported features will be updated with the release artifacts.
+The GGUF contains the converted chat template. Reduce `-ngl` or the context
+size if the available VRAM is smaller.
 
 ## Reproducibility and validation
 
-The locked upstream revision is `56f0584bb40578a2c33b1b40a08ccd17243ad710`.
-The direct BF16 conversion has completed and a deterministic CPU chat smoke
-test passed with the llama.cpp runtime. The final package will document the
-BF16 source hash, model-specific importance matrix hash, converter/runtime
-revision, quantization commands, artifact checksums, and load/generate
-validation. Detailed build and evaluation logs remain local; compact manifests
-and summaries will be added here as the release progresses.
+The locked upstream revision is
+`56f0584bb40578a2c33b1b40a08ccd17243ad710`. The BF16 source GGUF SHA256 is
+`10aae5451e1c0b6089b25f68cfe32cd32a4bdcb92e30008e083e6ba351f3cbd8`; the
+NeoHorse-specific combined imatrix SHA256 is
+`8c66b45b73826b0b00fdd973d308d8f81f7db0f8a3b97cf93bc7813d722d6515`.
+
+The package includes [SHA256SUMS.txt](SHA256SUMS.txt), the
+[quantization manifest](reproducibility/quantization-manifest.tsv), source and
+calibration manifests, the runtime version, and the copied upstream chat
+template. Detailed conversion, quantization, smoke, behavior, and fidelity
+logs remain local under `reports/neohorse-1-4b/` and are not uploaded.
 
 ## License and attribution
 
-NeoHorse-1-4B is released upstream under the Apache License 2.0. The upstream license and attribution notices will be included with the GGUF package.
+NeoHorse-1-4B is released upstream under the Apache License 2.0. The upstream
+license and attribution notices are included in `LICENSE`.
 
 These are community GGUF quantizations, not an official TokenRhythm release or endorsement.

SHA256SUMS.txt

@@ -0,0 +1,9 @@
+9b756eefd6ef534bf23ee3d62eb1b7e404ab4a21d8681a35791a0e39b76f17d0  NeoHorse-1-4B-Q8_0.gguf
+3af3f1981b8bc0a1fd205b011dc17cd02d5a479e89d35d77b8c20f5e6f0d70d5  NeoHorse-1-4B-Q6_K.gguf
+71b5e86dc409e1a5b17db34040f7df8dcd11688f27f6134d1c8595069ffe57fe  NeoHorse-1-4B-Q5_K_M.gguf
+5669ad48af5d54a29ed78aed77d2f2e2b921a8aca68c5482696bc8713cf8248b  NeoHorse-1-4B-Q4_K_M.gguf
+ef8b4386a54ec1d25024aeae5e5a1c976748bf2bd525c8cdc68edbec68a936f9  NeoHorse-1-4B-Q3_K_M.gguf
+834bef5dfdbbd2545b2fac9dd46a75c08beab8803a1636fd459cbe27729d8154  NeoHorse-1-4B-Q2_K.gguf
+a77f44a832a5978e4ce921c183cb07a80faa380fe4a8c22b0ba3a57699a34982  NeoHorse-1-4B-IQ2_XS.gguf
+eccb688df4b1c1387bee976db5972fe766d4a9007d3983132957587387efc0bb  NeoHorse-1-4B-IQ1_M.gguf
+9ab2feb77d3ac44677468d32142804df8346b41dc644ff95d0358d868a40aba0  NeoHorse-1-4B-Q1_0.gguf

reproducibility/calibration-manifest.md

@@ -0,0 +1,39 @@
+# Calibration manifest
+
+The importance matrix is model-specific and was collected from the BF16 GGUF.
+WikiText validation and test files were held out.
+
+| Input | SHA256 | Collection |
+|---|---|---|
+| `neohorse_en_vi_zh_code_agent.txt` | `235ed59cfc43beeaee7a0787273ff6586829b3ecf19189f4567fe57f64a03c3a` | context 512, requested 4 chunks, 3 collected |
+| `wiki.train.raw` | `6707892fa3788b5ab9ed78ab5ff37d9fe825f6011a2ad4fcd6a6d467f0e7da57` | context 4096, 96 chunks |
+| `wiki.valid.raw` | `4cd0f6876d07a413aa911261ff6d363c72d757d47f0fdd6015702014c89cb9c7` | hold-out only |
+| `wiki.test.raw` | `173c87a53759e0201f33e0ccf978e510c2042d7f2cb78229d9a50d79b9e7dd08` | hold-out only |
+
+Commands, with paths relative to the workspace root:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 llama-imatrix \
+  -m source-official-bf16-neohorse-1-4b/NeoHorse-1-4B-BF16.gguf \
+  -f calibration/neohorse-1-4b/neohorse_en_vi_zh_code_agent.txt \
+  -o calibration/neohorse-1-4b/NeoHorse-1-4B-custom.imatrix.gguf \
+  -c 512 -b 512 -ub 512 --chunks 4 --no-ppl --parse-special \
+  -ngl all --device CUDA0 --split-mode none --main-gpu 0 \
+  --fit on --fit-target 1024 -t 64 -tb 64
+
+CUDA_VISIBLE_DEVICES=0 llama-imatrix \
+  -m source-official-bf16-neohorse-1-4b/NeoHorse-1-4B-BF16.gguf \
+  -f calibration/wikitext-2-raw/wiki.train.raw \
+  -o calibration/neohorse-1-4b/NeoHorse-1-4B-wikitext-train.imatrix.gguf \
+  -c 4096 -b 512 -ub 512 --chunks 96 --no-ppl \
+  -ngl all --device CUDA0 --split-mode none --main-gpu 0 \
+  --fit on --fit-target 1024 -t 64 -tb 64
+
+llama-imatrix -m source-official-bf16-neohorse-1-4b/NeoHorse-1-4B-BF16.gguf \
+  --device none -ngl 0 \
+  --in-file calibration/neohorse-1-4b/NeoHorse-1-4B-custom.imatrix.gguf,calibration/neohorse-1-4b/NeoHorse-1-4B-wikitext-train.imatrix.gguf \
+  -o calibration/neohorse-1-4b/NeoHorse-1-4B-combined.imatrix.gguf
+```
+
+The combined imatrix SHA256 is
+`8c66b45b73826b0b00fdd973d308d8f81f7db0f8a3b97cf93bc7813d722d6515`.

reproducibility/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

reproducibility/manifest.md

@@ -0,0 +1,64 @@
+# NeoHorse-1-4B GGUF reproducibility manifest
+
+## Source
+
+- Upstream repository: `TokenRhythm/NeoHorse-1-4B`
+- Locked revision: `56f0584bb40578a2c33b1b40a08ccd17243ad710`
+- Source format: upstream BF16 safetensors
+- Architecture: `qwen3_5_text` / `Qwen3_5ForCausalLM`
+- Source BF16 GGUF: `NeoHorse-1-4B-BF16.gguf`
+- Source BF16 GGUF SHA256:
+  `10aae5451e1c0b6089b25f68cfe32cd32a4bdcb92e30008e083e6ba351f3cbd8`
+- Conversion used `--outtype bf16 --no-nextn`; the source index contains no
+  `mtp`/`nextn` tensors even though the upstream config advertises one MTP
+  layer.
+- The converter preserved the upstream text-only model and embedded its chat
+  template. It did not modify source weights or the upstream config.
+
+## Importance matrix
+
+- Combined imatrix: `NeoHorse-1-4B-combined.imatrix.gguf`
+- Combined imatrix SHA256:
+  `8c66b45b73826b0b00fdd973d308d8f81f7db0f8a3b97cf93bc7813d722d6515`
+- Input datasets: the model-specific multilingual/code/agent calibration text
+  and `wiki.train.raw`; neither WikiText validation nor test text was used.
+- The custom run used context 512 and four requested chunks (three available
+  chunks were collected). The WikiText run used context 4096 and 96 chunks.
+  The combined file reports 771 normalized chunks because the input matrices
+  use different context sizes.
+- Calibration input hashes and commands are in `calibration-manifest.md`.
+
+## Quantization
+
+Each published file was quantized directly from the BF16 GGUF with the Spark
+llama.cpp `llama-quantize` binary. `Q8_0` used no imatrix; every lower rung used
+the combined NeoHorse imatrix. The ladder order was:
+
+`Q8_0 -> Q6_K -> Q5_K_M -> Q4_K_M -> Q3_K_M -> Q2_K -> IQ2_XS -> IQ1_M -> Q1_0`
+
+Exact commands and artifact hashes are in `quantization-manifest.tsv` and
+`SHA256SUMS.txt`.
+
+## Runtime validation
+
+- Runtime repository: `tools-llama.cpp-spark`
+- Runtime commit: `4a3635c32fc9f044c2bde9ebeabf50c7e1ec5991`
+- Runtime version: `0.1.2-dev (build 1, commit 4a3635c)`
+- GPU validation: one NVIDIA A10M, process-local `CUDA_VISIBLE_DEVICES=0`,
+  `--device CUDA0`, `--split-mode none`, `--fit on`, fit target 1024 MiB,
+  Flash Attention on, context 4096, 64 threads.
+- BF16 and every published GGUF loaded and generated non-empty output using
+  the same profile. Low-bit smoke output is not a quality guarantee.
+- The pre-existing GPU 0 service was stopped only for the GPU jobs and
+  restored after validation; the GPU 1 service remained running.
+
+## Fidelity evaluation
+
+The compact results are in `quality-summary.tsv`. They average eight chunks
+from each of `wiki.test.raw` and `wiki.valid.raw`, context 4096, against the
+BF16 reference using the same runtime. These are next-token fidelity metrics,
+not a direct percentage of capabilities retained.
+
+Raw conversion, quantization, smoke, behavior, and fidelity logs remain local
+under `reports/neohorse-1-4b/` and are intentionally excluded from this public
+package.

reproducibility/quality-summary.tsv

@@ -0,0 +1,10 @@
+model	mean_kld	top1_percent	mean_ppl	mean_ppl_base	ppl_delta_percent	p_diff_rms_percent
+Q8_0	0.006291	98.065	8.651847	8.635307	0.192	2.041
+Q6_K	0.010910	96.751	8.707086	8.635307	0.831	2.881
+Q5_K_M	0.032008	94.498	8.782643	8.635307	1.706	4.631
+Q4_K_M	0.055031	91.909	8.750954	8.635307	1.339	5.920
+Q3_K_M	0.160972	85.173	9.625402	8.635307	11.466	10.610
+Q2_K	0.348854	76.007	9.733700	8.635307	12.720	16.212
+IQ2_XS	0.517311	70.662	11.433729	8.635307	32.407	20.489
+IQ1_M	1.303395	54.895	23.175517	8.635307	168.381	32.950
+Q1_0	12.339880	0.406	1588953.965748	8.635307	18400566.679	60.432

reproducibility/quantization-manifest.tsv

@@ -0,0 +1,10 @@
+file	type	bytes	sha256	imatrix	direct_source	runtime_smoke
+NeoHorse-1-4B-Q8_0.gguf	Q8_0	4482403104	9b756eefd6ef534bf23ee3d62eb1b7e404ab4a21d8681a35791a0e39b76f17d0	no	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q6_K.gguf	Q6_K	3464055872	3af3f1981b8bc0a1fd205b011dc17cd02d5a479e89d35d77b8c20f5e6f0d70d5	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q5_K_M.gguf	Q5_K_M	3074987072	71b5e86dc409e1a5b17db34040f7df8dcd11688f27f6134d1c8595069ffe57fe	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q4_K_M.gguf	Q4_K_M	2708804672	5669ad48af5d54a29ed78aed77d2f2e2b921a8aca68c5482696bc8713cf8248b	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q3_K_M.gguf	Q3_K_M	2262064192	ef8b4386a54ec1d25024aeae5e5a1c976748bf2bd525c8cdc68edbec68a936f9	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q2_K.gguf	Q2_K	1915470912	834bef5dfdbbd2545b2fac9dd46a75c08beab8803a1636fd459cbe27729d8154	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-IQ2_XS.gguf	IQ2_XS	1630594112	a77f44a832a5978e4ce921c183cb07a80faa380fe4a8c22b0ba3a57699a34982	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-IQ1_M.gguf	IQ1_M	1426418752	eccb688df4b1c1387bee976db5972fe766d4a9007d3983132957587387efc0bb	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass
+NeoHorse-1-4B-Q1_0.gguf	Q1_0	1038189632	9ab2feb77d3ac44677468d32142804df8346b41dc644ff95d0358d868a40aba0	NeoHorse-1-4B-combined.imatrix.gguf	NeoHorse-1-4B-BF16.gguf	pass

reproducibility/runtime-version.txt

@@ -0,0 +1,9 @@
+runtime_repo=tools-llama.cpp-spark
+runtime_commit=4a3635c32fc9f044c2bde9ebeabf50c7e1ec5991
+runtime_version=0.1.2-dev (build 1, commit 4a3635c)
+converter=tools-llama.cpp-spark/convert_hf_to_gguf.py
+converter_options=--outtype bf16 --no-nextn --model-name NeoHorse-1-4B
+quantizer=tools-llama.cpp-spark/build/bin/llama-quantize
+perplexity=tools-llama.cpp-spark/build/bin/llama-perplexity
+chat_runtime=tools-llama.cpp-spark/build/bin/llama-cli
+gpu_validation=NVIDIA A10M physical GPU 0; CUDA_VISIBLE_DEVICES=0; GPU 1 service left running

reproducibility/source-manifest.md

@@ -0,0 +1,17 @@
+# Source manifest
+
+Upstream revision: `56f0584bb40578a2c33b1b40a08ccd17243ad710`
+
+| File | SHA256 |
+|---|---|
+| `model-00001-of-00002.safetensors` | `d7f64e7920d47e76bac3296b58e90e7902e3c686f4fb4891303ab832c898bcc3` |
+| `model-00002-of-00002.safetensors` | `36292efc411eb9580ada5f14a792712dd8a4d9e9657812d070d2d0435a64b9f5` |
+| `model.safetensors.index.json` | `315b36f21026e93b74552e5755ec9a52ab6232de8c54f0ecb34fc00a50c7a20a` |
+| `config.json` | `e80a48542cf4dfbd60f19d3fea5a8a6f89e1c7fa65c992fea572aeeaa906b29c` |
+| `tokenizer.json` | `5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42` |
+| `tokenizer_config.json` | `316230d6a809701f4db5ea8f8fc862bc3a6f3229c937c174e674ff3ca0a64ac8` |
+| `chat_template.jinja` | `a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715` |
+| `LICENSE` | `c2c9ab53cedffada40bb3d5ccf733f931fdabb3ee0759e7a7e1eb30d42112df9` |
+
+Converted BF16 GGUF SHA256:
+`10aae5451e1c0b6089b25f68cfe32cd32a4bdcb92e30008e083e6ba351f3cbd8`.

reproducibility/summarize_fidelity.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""Summarize NeoHorse GGUF next-token fidelity across fixed hold-outs."""
+
+from __future__ import annotations
+
+import argparse
+import re
+from pathlib import Path
+
+
+FLOAT = r"[-+]?\d+(?:\.\d+)?(?:[eE][-+]?\d+)?"
+
+
+def value(text: str, pattern: str) -> float:
+    matches = re.findall(pattern, text)
+    if not matches:
+        raise ValueError(f"pattern not found: {pattern}")
+    return float(matches[-1])
+
+
+def parse_log(path: Path) -> dict[str, float]:
+    text = path.read_text(errors="replace")
+    return {
+        "mean_kld": value(text, rf"Mean\s+KLD:\s*({FLOAT})"),
+        "top1_percent": value(text, rf"Same top p:\s*({FLOAT})"),
+        "mean_ppl": value(text, rf"Mean PPL\(Q\)\s*:\s*({FLOAT})"),
+        "mean_ppl_base": value(text, rf"Mean PPL\(base\)\s*:\s*({FLOAT})"),
+        "p_diff_rms_percent": value(
+            text, rf"RMS Δp\s*:\s*({FLOAT})(?:\s*±\s*{FLOAT})?\s*%"
+        ),
+    }
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("logdir", type=Path)
+    parser.add_argument("output", type=Path)
+    parser.add_argument("--chunks", type=int, default=8)
+    parser.add_argument("--splits", nargs="+", default=["test", "valid"])
+    parser.add_argument("--quants", nargs="+", required=True)
+    args = parser.parse_args()
+
+    args.output.parent.mkdir(parents=True, exist_ok=True)
+    with args.output.open("w", encoding="utf-8") as handle:
+        handle.write(
+            "model\tmean_kld\ttop1_percent\tmean_ppl\tmean_ppl_base\t"
+            "ppl_delta_percent\tp_diff_rms_percent\n"
+        )
+        for quant in args.quants:
+            records = [
+                parse_log(args.logdir / f"{quant}-{split}-chunks{args.chunks}.log")
+                for split in args.splits
+            ]
+            mean_kld = sum(r["mean_kld"] for r in records) / len(records)
+            top1 = sum(r["top1_percent"] for r in records) / len(records)
+            mean_ppl = sum(r["mean_ppl"] for r in records) / len(records)
+            mean_base = sum(r["mean_ppl_base"] for r in records) / len(records)
+            rms = sum(r["p_diff_rms_percent"] for r in records) / len(records)
+            delta = (mean_ppl / mean_base - 1.0) * 100.0
+            handle.write(
+                "\t".join(
+                    (
+                        quant,
+                        f"{mean_kld:.6f}",
+                        f"{top1:.3f}",
+                        f"{mean_ppl:.6f}",
+                        f"{mean_base:.6f}",
+                        f"{delta:.3f}",
+                        f"{rms:.3f}",
+                    )
+                )
+                + "\n"
+            )
+    print(f"wrote {args.output} ({len(args.quants)} models)")
+
+
+if __name__ == "__main__":
+    main()